Direct Ingestion

The proxy is the easiest way to log tool calls, but some agents connect directly to MCP servers and can't be rerouted. For these cases, use the direct ingestion endpoint to push tool call records to MCPHub manually.

When to use direct ingestion

Use it when:

• Your agent manages its own MCP connections and you can't change the transport layer
• You want to log tool calls from a legacy system that predates MCPHub
• You're running MCP servers in an environment where proxying isn't feasible (e.g. local dev tools, sandboxed environments)

Endpoint

POST /api/v1/tool-calls

Requires JWT or API key auth.

Request schema

{
  "server_id": "uuid-of-registered-server",
  "tool_name": "read_file",
  "caller_agent": "my-agent-v2",
  "input_tokens": 120,
  "output_tokens": 340,
  "duration_ms": 245,
  "status": "success",
  "error_message": null,
  "output_size_bytes": 1380
}

Example (Python)

import httpx
 
httpx.post(
    "https://mcphub.aniruddha.fyi/api/v1/tool-calls",
    headers={"X-API-Key": "mhk_your_key"},
    json={
        "server_id": "your-server-uuid",
        "tool_name": "search_files",
        "caller_agent": "my-pipeline",
        "duration_ms": 312,
        "status": "success",
        "output_size_bytes": 4096,
    },
)

What you get

Directly ingested tool calls appear in:

• Tool Calls audit log (same as proxy-logged calls)
• Analytics — top tools, volume heatmap, error rates
• Alert evaluation — alert rules fire on ingested data too

The only difference from proxy logging is that directly ingested calls don't have an MCPHub-generated request ID, so there's no link back to the raw request.